State departments of transportation (DOTs) use different pavement surfaces, such as open-graded friction courses (OGFC), seal coats, and chip seals. Raveling (loss of aggregates) is a predominant pavement distress that impacts the safety and functionality of the pavement surface. Some state DOTs have classified raveling into different severity levels to determine the appropriate pavement preservation actions. However, current practices for the visual inspection of the severity levels of raveling are time-consuming, labor-intensive, and, most importantly, subjective.
NCHRP IDEA 20-30/IDEA 163, "Development of an Asphalt Pavement Raveling Detection Algorithm using Emerging 3-D Laser Technology and Macrotexture Analysis," successfully developed automatic raveling detection and classification algorithms using three-dimensional (3D) pavement surface condition data, macro-texture analysis, and machine learning (ML) modeling. Different ML models were critically evaluated, and automatic raveling detection and classification algorithms were developed. The algorithms use the 3D pavement surface data already collected by state DOTs for the evaluation of cracking and rutting in pavements, so additional data collection effort is not needed. The output from the automatic raveling detection and classification algorithm is the severity level (severe/3, medium/2, and low/1) based on the 3D pavement images collected at different image sizes (e.g., 5-meter or 8-meter intervals) as specified by the different 3D sensing systems used.
Based on the process described in NCHRP IDEA 20-30/IDEA 163, Florida DOT developed its own algorithm for classifying pavement raveling. The Florida DOT algorithm, which was written in Python, uses a random forest classifier, and the algorithm development included a training module where human raters looked at images and classified them. This algorithm can also read images, process them, and classify raveling by severity.
The objective of this research is to develop guidelines for implementing the automatic raveling detection and classification algorithms developed in NCHRP IDEA 20-30/IDEA 163 in available programming languages or commercial software, such as Python or MATLAB. State DOTs should be able to use the products to calibrate and refine the algorithms for future use, which may include the use of different severity classifications and surface types, as well as incorporate the algorithms in their pavement management system or use the algorithms to make maintenance and rehabilitation decisions.
Accomplishment of the project objective will require at least the following tasks.
Task 1. Review the final report and algorithm from NCHRP IDEA 20-30/IDEA 163. Conduct a comprehensive survey of the tools and software that state DOTs are most likely to use in order to efficiently implement automatic raveling detection and classification algorithms on a network level consisting of millions of 3D images. Image quality, image format, and image/data processing techniques critical to accurately implementing the raveling classification method should be considered. The research team should review algorithms developed by Florida DOT and other state DOTs that have implemented the results of NCHRP IDEA 20-30/IDEA 163.
Deliverables: Memorandum summarizing image properties, processing techniques, tools, and software.
Task 2. Develop criteria to determine a state DOT’s eligibility to participate in pilot studies (see Task 4) and training (see Task 5). The criteria should include, at a minimum, the desired resolution and format of the raw image files a state DOT would provide, data qualifications, state DOT data processing capability, raveling type and severity, the length of roadway, and representation from all four AASHTO regions. Propose state DOTs that meet the criteria for inclusion in the pilot studies. NCHRP approval of the participant state DOTs is required before work on Task 4 may begin.
Deliverables: Report with a list of criteria as well as a list of potential participating state DOTs.
Task 3. Develop guidelines for implementing the algorithms, which should be vendor neutral and applicable in any jurisdiction. At a minimum, this will include reviewing the automatic raveling detection and classification algorithms, which were developed for OGFC, for algorithmic bias; identifying the steps needed to refine and validate the algorithms for other surfaces; and proposing how the algorithms could be made more efficient.
Deliverables: Guidelines for implementing the automatic raveling detection and classification algorithms.
Task 4. Using the guidelines from Task 3, perform pilot studies of the automatic raveling detection and classification algorithms with at least two host agencies. The following subtasks will need to be performed for each host agency:
- Customize the data processing tool to load and process a sufficiently large quantity of 3D pavement image data and to store classification outcomes in a relational database with a standardized data format.
- Customize the outcome quality checking tool for use by a host agency.
- Customize how the tool aggregates automatically generated image-based classified raveling outcomes into a data management system appropriate for state DOTs, which could be used for pavement management or quality assurance/quality control reporting.
Deliverables: Detailed reports of pilot studies with two state DOTs.
Task 5. Conduct virtual training with at least five additional state DOTs on how to implement the automatic raveling detection and classification algorithms, tools, and functions into a state DOT’s data collection and management processes. Virtual site visits and training will be conducted using sample data provided from the pilot sites in Task 4. This task will include preparing training materials; gathering feedback from participating state DOTs to identify common practices and streamlining the process for other state DOTs' implementation; and refining the training materials based on experience conducting training and participant feedback. The online training materials developed will be posted on the project page.
Deliverables: Training with at least five state DOTs and training materials.
Task 6. Prepare outreach materials for the guidelines for implementing the automatic raveling detection and classification algorithms and the training materials, which may include briefing materials for senior management, flyers, brochures, and videos. Disseminate results through a TRB webinar, along with presentations at appropriate AASHTO and TRB meetings.
Deliverables: Outreach materials.
Task 7. Prepare a conduct of research report that documents the activities of this project, describes experiences with pilot studies and training, and captures lessons learned.
Deliverable: Final project report.
STATUS: A research agency has been selected for the project. The contracting process is underway.